Skip to content
Bencher

Models · Swiss AI Initiative (EPFL, ETH Zurich, CSCS) · Out sinceReleased 24 Jul 2026

Apertus 1.5 70B

Only from the makerNot on OpenRouterAvailable nowGenerally availableCan run on your own serversOpen weights

Apertus 1.5 70B is made by Swiss AI Initiative (EPFL, ETH Zurich, CSCS). We don't have enough test results yet to rank it. Its makers have published it, so you can run it on your own servers.

Swiss fully-open model (EPFL/ETH/CSCS): continued pretraining of Apertus 1.0 (+2T tokens), adds image input, thinking mode and 262,144-token context. Apache 2.0 but HF download is gated behind accepting the Apertus 1.5 Acceptable Use Policy and sharing contact details. HF safetensors metadata: ~72.0B params. Card itself is gated (not read).

Writing & creativity
—
Research & analysis
—
Coding
—
Price
Price unknown— / —
Price per 1M (blended)Blended / 1M
—
MemoryContext
262K
Longest answerMax output
—
Test resultsResults
1 (1 independent1 indep.)
Out sinceReleased
24 Jul 2026
Made byVendor
Swiss AI Initiative (EPFL, ETH Zurich, CSCS)
UnderstandsInputs
text, image

Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.

Thinking levels

Reasoning settings

No reasoningDefault

Read more

Links

Run it yourself

Self-hosting

On your own servers,Licence, sizeunder your own control.and hardware.

Compare all open models →All open-weights models →

LicenceLicence
Apache-2.0 (+ Apertus 1.5 Acceptable Use Policy at download gate)
OK for business useCommercial use
Yes
SizeParams (total / active)
72B
Can be trained furtherBase model
No

Hardware you'd need

Hardware tier & memory

Runs on one server graphics card.

≤ 80 GB at 4-bit (one H100/H200). Weights ≈ 76 GB at 8-bit, 40 GB at 4-bit (+10–30% for KV cache).

Licence conditions: HF access gate requires accepting Apache 2.0 + the Apertus 1.5 AUP and sharing company/institutional email (privacy policy applies). AUP contents not reviewed.

Restrictions: HF access gate requires accepting Apache 2.0 + the Apertus 1.5 AUP and sharing company/institutional email (privacy policy applies). AUP contents not reviewed.

Languages: Languages: Not verified for 1.5 (card gated). EuroEval Swedish rank score 2.11.

Training it further: Fine-tuning: Not verifiable - model card gated. Fully-open project (training data/recipes published for Apertus 1.0).

Quantisations: fp8 (community: onprem-ai), gguf (community).

Download (Hugging Face)Weights on Hugging Face ↗huggingface.co ↗apertus-ai.org ↗publicai.co ↗

Every test result

Every result

All the numbers,with where they came from.

Every result we've found for this model, with who measured it and a link to where we read it.

All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.

All results for this model
NotesNotes
EuroEval Swedish (generative)2.1Defaultswiss-ai/Apertus-v1.5-70B (val)Independent testIndependentEuroEval (Alexandra Institute) ↗29 Sep 2026—rank tier 10; ±0.10; lower is better; SweDN 38.37, Skolprov 34.81, Swedish facts 29.66, ScaLA-sv 56.88